I have noticed this myself, especially with the Cree LEDs -- the "warmer" light is emitted to the sides, as the blue light coming from the die passes through a greater amount of phosphor materials that the light that exits head-on.
This produces the effect of reflector-based lights having a "warmer" hotspot, with cooler spill, which IMO is the best way to do it, as the majority of the blue receptors in the eye are in the peripheral vision, with the central vision almost entirely composed of red/green receptors. Of course, for area illumination (not a flashlight) that creates problems, however, area illumination is still better off with typical fluorescent lighting, IMO.
Interestingly, they mentioned near-UV based LEDs with separate RGB phosphors. This setup makes very little sense IMO, one of the advantages of using a blue + yellow, or blue + separate red/green phosphors is that there is NO UV content, which would be important for something like museum lighting etc.
I personally have found that the best color rendering from LEDs so far has been achieved by finding LEDs with a "greenish" cast to them (stronger phosphor emissions, weaker blue spike), then adding separate red emitters to the mix. Even this does not solve the problem of phosphor variations from one part to another, which are apparently very difficult to control.
It seems like some sort of RGB emitter with three separate dice will eventually be the best way to solve all the problems mentioned so far -- namely color rendering, and control of tint. I believe Cree submitted a patent application a while back for a RGB emitter with the separate dice stacked one on top of the other, as opposed to side-by-side. An RGB chip that didn't have color mixing problem would really be great, and woudl allow a lot more freedom both in flashlights, and in general use.